Research map / Detection and segmentation / CNN features
Decoders Matter for Semantic Segmentation: Data-Dependent Decoding Enables Flexible Feature Aggregation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace bilinear upsampling with DUpsampling at output stride 32 (ablation) | ResNet-50 segmentation decoder · PASCAL VOC val | mIOU (%): 70.77 -> 72.09 | Helped |
| Replace bilinear upsampling with DUpsampling at output stride 16 (ablation) | ResNet-50 segmentation decoder · PASCAL VOC val | mIOU (%): 72.15 -> 73.15 | Helped |
| Fuse conv1_3 features instead of using no low-level features (ablation) | Proposed ResNet-50 decoder · PASCAL VOC val | mIOU (%): 73.15 -> 72.70 | Hurt |
| Fuse b1u2c3 features instead of using no low-level features (ablation) | Proposed ResNet-50 decoder · PASCAL VOC val | mIOU (%): 73.15 -> 74.03 | Helped |
| Fuse b3u6c3 features instead of using no low-level features (ablation) | Proposed ResNet-50 decoder · PASCAL VOC val | mIOU (%): 73.15 -> 73.43 | Helped |
| Fuse b1u2c3 and b3u6c3 features instead of using no low-level features (ablation) | Proposed ResNet-50 decoder · PASCAL VOC val | mIOU (%): 73.15 -> 73.82 | Helped |
| Fuse conv1_3 and b3u6c3 features instead of using no low-level features (ablation) | Proposed ResNet-50 decoder · PASCAL VOC val | mIOU (%): 73.15 -> 74.20 | Helped |
| Replace vanilla softmax with adaptive-temperature softmax (ablation) | DUpsampling framework · PASCAL VOC val | mIOU (%): 69.81% -> 73.15% | Helped |
| Replace the vanilla decoder with the proposed decoder using the same b1u2c3 features (ablation) | ResNet-50 decoder · PASCAL VOC val | mIOU (%): 73.26% -> 74.03% | Helped |
| Use conv1_3 + b3u6c3 feature fusion with the proposed decoder (ablation) | ResNet-50 decoder · PASCAL VOC val | mIOU (%): 74.03% -> 74.20% | Helped |
| Replace the vanilla decoder with the proposed decoder using efb2u1c2 features (ablation) | Xception-65 decoder · PASCAL VOC val | mIOU (%): 78.70% -> 79.09% | Helped |
| Use mfb1u16c3 rather than efb2u1c2 feature fusion with the proposed decoder (ablation) | Xception-65 decoder · PASCAL VOC val | mIOU (%): 79.09% -> 79.67% | Helped |
Closest papers by meaning
- SED: A Simple Encoder-Decoder for Open-Vocabulary Semantic Segmentation (CVPR 2024 · similarity 0.90)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.88)
- Open-Vocabulary Semantic Segmentation with Image Embedding Balancing (CVPR 2024 · similarity 0.87)
- MaSS13K: A Matting-level Semantic Segmentation Benchmark (CVPR 2025 · similarity 0.87)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.87)
- DPSeg: Dual-Prompt Cost Volume Learning for Open-Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.